Nobel Numbers: Time-Dependent Centrality Measures on Co-Authorship Graphs
Total Page：16
File Type：pdf, Size：1020Kb
C. Fields, p. 1 of 27 “This is a preprint of an article accepted for publication in Journal of the Association for Information Science and Technology copyright © 2015 (Association for Information Science and Technology)” Nobel numbers: Time-dependent centrality measures on co-authorship graphs Chris Fields 528 Zinnia Court Sonoma, CA 95476, USA [email protected] Abstract A time-dependent centrality metric for disciplinary co-authorship graphs, the “Nobel number” for a discipline, is introduced. A researcher's Nobel number for a given discipline in a given year is defined as the researcher's average co-authorship distance to that discipline's Nobel laureates in that year. Plotting Nobel numbers over several decades provides a quantitative as well as visual indication of a researcher's proximity to the intuitive “center” of a discipline as defined by recognized scientific achievement. It is shown that the Nobel number distributions for physics of several researchers both within and outside of physics are surprisingly flat over the five-decade span from 1951 to 2000. A model in which Nobel laureates are typically connected by short co-authorship paths both intergenerationally and between subdisciplines reproduces such flat Nobel number distributions. Keywords: Betweenness centrality; Biomedicine; Cross-disciplinary brokerage; Erdős numbers; Interdisciplinarity; Physics Introduction What does it mean to say that a researcher is “central” to a discipline? In particular, what does it mean to say that a researcher is central to the social network represented by the discipline's co-authorship graph? Formal, graph-theoretic measures of centrality provide some candidate answers. One can identify the researchers with the largest numbers of co-authors, who have “degree centrality” in graph- theoretic terms (for definitions of relevant terms and concepts of graph theory, see Börner, Sanyal and Vespignani, 2007; Diestel, 2010). Alternatively, one can calculate the length of longest minimal path between any two researchers; this length is the “diameter” of the co-authorship graph. Any researcher C. Fields, p. 2 of 27 separated by no more than half of the graph diameter from any other researcher has “distance centrality” within the graph. A third option is to identify those researchers through whom the largest numbers of minimal paths connecting other researchers flow; they have “betweenness centrality.” These three formal measures of centrality do not pick out the same researchers as “central” unless the co-authorship graph is highly symmetrical (see Fields, 2014, Fig. 2 for an example graph in which the three measures coincide and another in which they do not), and their meaningfulness as measures of centrality in co-authorship graphs remains a topic of considerable debate (e.g. Freeman, 1978/79; Borgatti and Everett, 2006; Landherr, Friedl and Heidemann, 2010). Minor members of large collaborations, for example, may have many co-authors and hence high degree centrality, otherwise- unremarkable “weak links” between subdisciplines may have high betweenness centrality, and distance centrality reflects academic prominence within a discipline only if the most prominent individuals, and no others, occupy the metric center of the discipline's co-authorship graph. Intuitively and informally, the centrality of a researcher within a discipline is not determined by his or her position in the co-authorship graph, but rather by relevant scientific accomplishments. Well- known, well-respected intellectual leaders, who are often also well-funded, highly-productive political leaders, are widely acknowledged to be “central” players in any discipline. The recognitions and honors that disciplines bestow, such as Nobel Prizes, Fields Medals or Abel Prizes in mathematics, the Catherine Wolfe Bruce Gold Medals in astronomy or A. M. Turing Awards or John von Neumann Medals in computer science, are generally awarded to such recognized leaders. It seems reasonable, therefore, to regard “centrality” within a co-authorship graph not as a formal, graph-theoretic notion but rather as informally defined by recognized scientific accomplishments, and to seek metrics for co- authorship centrality that honor this informal definition. As Nobel prizes are awarded annually, the Nobel laureates of a discipline provide a time-dependent marker of the informal “center” of that discipline as defined by recognized scientific achievement. Note that “recognized” here means recognized by the award of a Nobel prize, not recognized at the time that the relevant work was done or published, a discrepancy perhaps most notable in the case of Barbara McClintock (Physiology or Medicine, 1983),1 whose work was largely dismissed when first reported (Keller, 1983). Research work that is regarded as “central” at the time that a Nobel prize is awarded may, in other words, not have been regarded as “central” or even as valid when the work was performed; similarly, researchers who are regarded as “central” when a Nobel prize is awarded may have been regarded as marginal when they performed their Nobel-prize-winning work. Here I propose a time-dependent centrality metric for disciplinary co-authorship graphs, the “Nobel number” for a discipline, that reflects this informal, community-recognition-based notion of centrality. Nobel numbers measure the average distance in co-authorship units – i.e. the average of the minimal co-authorship path lengths – from any researcher to that discipline's Nobel laureates for a given year. Metrics based on other community-awarded prizes, such as Fields Medals or Turing Awards, can be analogously defined. To illustrate the utility and informativeness of this kind of metric, I consider the Nobel laureates in Physics in the latter half of the 20th century, i.e. from 1951 to 2000. This is the period during which high-energy physics emerged as a distinct and financially dominant subdiscipline within physics; it also saw the development of high-speed computers and “big science” collaborations, the emergence of solid-state physics, laser physics, observational astrophysics and quantum 1 Nobel laureates are indicated throughout by an award date. The award category is included for laureates in disciplines other than physics. C. Fields, p. 3 of 27 information theory, and technological developments including space travel, precision time measurement and the ability to manipulate single atoms. The time-dependent Nobel number NP(t) of any researcher with respect to this collection of 108 Nobel laureates in Physics would be expected to provide a natural, time-dependent measure of his or her distance from the recognized intellectual “center” of physics during this period of rapid growth and development. One might expect the Nobel number distribution for any individual with respect to any discipline awarding Nobel prizes to exhibit significant temporal structure, e.g. to have broad maxima and minima over the time range examined that correlate with the individual's subdiscipline, academic lineage, or some other variable(s). I show that, at least for physics during the time period examined, this does not appear to be the case. Even the Nobel number distributions for physics of biomedical researchers – here, Nobel laureates in Physiology of Medicine from 1991 to 2010 – are remarkably flat. I interpret this somewhat surprising result in terms of the high degree of both intergenerational and cross- subdisciplinary co-authorship connections among Nobel laureates in physics that is evident even in small samples from the complete co-authorship graph. Methods and Data Names and specializations of all Nobel laureates in Physics from 1951 to 2000 were obtained from Nobelprize.org.2 Co-authorship data were obtained from Google Scholar™ as described (Fields, 2015 a); co-authorship data were not filtered or otherwise restricted on the basis of the date of the co- authored publication. Paths from laureates traversing other authors known to be near either cross- disciplinary brokers or other Nobel laureates were followed preferentially. This is a tractable, greedy search procedure that produces upper limits on the minimal co-authorship path lengths from laureates to either other laureates or brokers; more exhaustive, and in particular non-heuristic, search procedures may produce smaller path lengths in some cases. “Cross-disciplinary brokers” were defined as individuals who have published co-authored papers both in physics and in at least one of the 15 other Klavans and Boyack (2009) consensus disciplines. Disciplinary assignments of papers were determined from the title or abstract, or if necessary, by reading the paper in its entirety. Only primary and secondary research papers, review articles, research-based science-policy papers and scholarly books were included in the co-authorship analysis; otherwise-unpublished technical reports, textbooks, joint editing of collections, and editorial or opinion pieces were not included. Where necessary, authors with similar names were disambiguated by tracing their histories of institutional appointments. The time-dependent Nobel number NP(t) for physics of a given researcher is defined as the average, for the year t, of the minimal co-authorship path lengths from that researcher to that year's Nobel laureates in Physics. The time-dependent Nobel number NM(t) for biomedicine is defined similarly, for each year's Nobel laureates in Physiology or Medicine. All values of NP(t) or NM(t) quoted or displayed here were computed from upper limits on minimal